Date of Graduation


Document Type


Degree Type



Statler College of Engineering and Mineral Resources


Lane Department of Computer Science and Electrical Engineering

Committee Chair

Guodong Guo

Committee Co-Chair

Donald A. Adjeroh

Committee Member

Cun-Quan Zhang


Human and animal behavior understanding is an important yet challenging task in computer vision. It has a variety of real-world applications including human computer interaction (HCI), video surveillance, pharmacology, genetics, etc. We first present an evaluation of spatiotemporal interest point features (STIPs) for depth-based human action recognition, and then propose a framework call TriViews for 3D human action recognition with RGB-D data. Finally, we investigate a new approach for animal behavior recognition based on tracking, video content extraction and data fusion.;STIPs features are widely used with good performance for action recognition using the visible light videos. Recently, with the advance of depth imaging technology, a new modality has appeared for human action recognition. It is important to assess the performance and usefulness of the STIPs features for action analysis on the new modality of 3D depth map. Three detectors and six descriptors are combined to form various STIPs features in this thesis. Experiments are conducted on four challenging depth datasets.;We present an effective framework called TriViews to utilize 3D information for human action recognition. It projects the 3D depth maps into three views, i.e., front, side, and top views. Under this framework, five features are extracted from each view, separately. Then the three views are combined to derive a complete description of the 3D data. The five features characterize action patterns from different aspects, among which the top three best features are selected and fused based on a probabilistic fusion approach (PFA). We evaluate the proposed framework on three challenging depth action datasets. The experimental results show that the proposed TriViews framework achieves the most accurate results for depth-based action recognition, better than the state-of-the-art methods on all three databases.;Compared to human actions, animal behaviors exhibit some different characteristics. For example, animal body is much less expressive than human body, so some visual features and frameworks which are widely used for human action representation, cannot work well for animals. We investigate two features for mice behavior recognition, i.e., sparse and dense trajectory features. Sparse trajectory feature relies on tracking heavily. If tracking fails, the performance of sparse trajectory feature may deteriorate. In contrast, dense trajectory features are much more robust without relying on the tracking, thus the integration of these two features could be of practical significance. A fusion approach is proposed for mice behavior recognition. Experimental results on two public databases show that the integration of sparse and dense trajectory features can improve the recognition performance.